Polygraf’s Post

Stop asking ChatGPT to 'try again'. Use structured outputs. Build your own agent. Start getting reliable one-shot outputs without much effort. If you're interested in how I build custom AI Agents for me and my team at MJV in less than 15 minutes, comment interested below.

I found Substrata, an AI tool that analyzes email negotiations and is aimed at dealmakers. I conducted an experiment where I did a fake negotiation with ChatGPT to see how the tool works. I hope you enjoy it! Links are in the comments!

"Shorter." "Even shorter." "No, I mean like 'CEO' short." These are the frequent prompts I give to ChatGPT when refining my messages. If you're like me, suffering from chronic over-elaboration, it might be time to consult your AI of choice and find out if Brevity is right for you. New users of Brevity have reported side effects that include a sense of frustration from not expressing all your thoughts, a tinge of doubt about the adequacy of your evidence, and a peculiar unease with the newfound silence and space in your communication. #brevity #overexplained #leadership #communication #ai #artificialintelligence #lesswords

Every AI power user that I know has switched from ChatGPT to Claude. But you need to know how to write long, complex prompts to get the most out of it. You need to use: 1. XML tags for clarity 2. Chain of thought for 39% better responses 3. Output and input pairs as examples 📌 Don't miss my new guide with 7 advanced techniques for prompting Claude: https://lnkd.in/ggc5yA6i

Peter Yang excellent post. In my experience, writing long prompts using the 5 keys gives me the best and most relevant responses. The 5 keys are - Role, Input, Instructions, Format, and Example. I have also found that order matters since the LLM (like humans) can get forgetful about information in the middle. Also, I am curious about your philosophy / intent when you write prompts. When I write prompts (and I know this is going to be controversial) my goal is to get high quality response I want in a single shot / minimal sparring. I want the prompt to be reusable - write it once, change inputs, and use it again and again. And the ultimate test - it has to save me upto 80% of time. If not I keep working on the prompt till I get there. How do you approach it?

Want to improve your ChatGPT answers by 50%?  New research has identified 26 strategies to improve your prompts... Turns out you don't need to say "please" or "thank you" to ChatGPT, but... You do need to be specific, intentional, and to the point... I am curious, how are you using ChatGPT in your work or in your personal life?  Link to the original research below and hat tip to the Superhuman AI newsletter for the great find 👇

Are you uploading files to ChatGPT and still getting half-baked responses? Meet RAG, the secret ingredient you've been missing. It's like giving ChatGPT a personal library instead of a book with missing pages. Embrace RAG through custom GPTs—you'll power up your business with tailored AI insights more accurate than ever.

One of my top ChatGPT hacks: Fork your GPTs. Don’t make your GPT a do it all machine. Don’t “kill two birds with one stone.” Niche down. Ex: I was helping a real estate leader on their lead gen (my neighbor’s mom, actually). She was using the same GPT to write all of her outreach emails. Now she has 3 different GPTs, based on customer personas. It took about 20min to build and test. Reply rate is already way up. If you want more quick tips that will improve your AI collabs, you must come to my prompting session with Conor Grennan and Cher Jones on July 31: https://lnkd.in/ecF73Xsp

Sometimes I get thoughts and ideas that are not always worth sharing, but I try to share them anyways. You never know when a productive conversation will come up when your mind is going creative. 💡 Building products is a process that involves a lot of: - people telling you you're crazy/wrong - bad ideas - failures - self criticism - radical candor - asking "why?" - hitting walls - breaking walls - pivoting - testing people's limiting beliefs - testing YOUR limiting beliefs Today, I had some ideas...I shard them with Marian Ignev as together with our team we're building products for people that are building the products of tomorrow. Those are AI-first products. Hell...I'll call them even the products of today, as AI is changing so fast, that what you think today is not what you need to do tomorrow! 🤯 So I'm also asking you - are you sure that what you're building today is what the market will need tomorrow? Or today it's already outdated? Be honest. Your product's success depends on that 🤔

ChatGPT is insanely helpful But the trolls say it is useless. Because the trolls use the worst prompt frameworks Here is a cheatsheet on the ChatGPT Role, task, and target audience prompt engineering framework: Save this post for future reference --- How to create the prompt: You are an [mention role]. Do [mention task with context and problem you’re facing] for [target audience]. --- Repost this post to be the first one to share with your network Learn AI for FREE: https://drp.li/GExM9 ---